AMD Threadripper Socket TR4 Up Close and Personal

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
54,075
So after our adventures with destroying the socket on our Gigabyte Aorus Gaming 7, I thought I would try to get some good pictures of the TR4 socket to show you just how jam-packed the socket is with all of its 4,096 pins. This is a full resolution image, 4000x3000, optimized to 2.1MB.

Picture on the news page.

Also worth mentioning again, as some folks do not seem to get my message. I did not damage the TR4 socket while installing the Threadripper CPU, that process was excellent in my opinion and shows how much thought and design went into its production. I damaged the TR4 pins when reinstalling the cover that protects the pins during shipment and install. This was totally my fault and none of the socket's design or implementation. However, all that does not negate the fact that these TR4 pins are incredibly delicate. So again, my message is to give warning to our readers that will be handling these processors.
 

Riccochet

Fully [H]
Joined
Apr 11, 2007
Messages
27,921
Kyle, in your opinion is it difficult to reinstall the protective cover? In case someone wanted to re-sell a mobo. Or would it be easier to just reinstall the clear plastic cassette and forgo the cover?
 

gxp500

Gawd
Joined
Mar 4, 2015
Messages
865
It looks like getting any kind of grip on that cover is the problem, if there would of been some pull tab on top to firmly grab it, it would solve the issue.
 

Advil

2[H]4U
Joined
Jul 16, 2004
Messages
2,091
I understand why they did it that way, but it underscores just how badly we need a new mounting method in the industry. Optical interconnects or something. I know you messed up, you admit you did and the motherboard maker has nothing to do with it. It's AMD's standard. But that really is kind of lunacy on pin count and square inches.

We've had good luck since the industry started to put the pins on the motherboards. None of us knew how that would work out in the beginning. It stopped the rampant pin damage on expensive CPUs. And mostly in enthusiast situations the motherboard is a less expensive item to destroy than the CPU.

But that socket is so large only people who trust themselves, have steady hands and know PC builds inside out should be screwing with that socket at all. (You certainly do, and an accident still happened) Even small debris contaminates in the pin area are an issue with that thing. You really wouldn't want to leave that pad exposed for very long. It's like a Murphy's Law Magnet. There's no way he can resist causing everything to go wrong. :p
 

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
54,075
Kyle, in your opinion is it difficult to reinstall the protective cover? In case someone wanted to re-sell a mobo. Or would it be easier to just reinstall the clear plastic cassette and forgo the cover?
No, it is not difficult, I just fat-fingered it, but I hardly "jammed" it down onto the pins either. You just needed to touch and have them snag something. Did this twice. Again, DO NOT TOUCH the pins at all.
 

ir0nw0lf

Supreme [H]ardness
Joined
Feb 7, 2003
Messages
6,404
1tl8nf.jpg
1tl27x.jpg

1tl8nf
 
Last edited:

Pusher of Buttons

[H]ard|Gawd
Joined
Dec 6, 2016
Messages
1,924
Contest to give away the damaged board for someone to try in desperation to get the pins straightened?
 

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
54,075
I thought this was [H]ardForum? Put the damn board out of its misery with the 50 cal:D
Hehe, sadly this boards are hardly thowaways in terms of cost. It is on its way back to Gigabyte.
 

Wolfie

Limp Gawd
Joined
Jun 2, 2006
Messages
501
How many hours have I spent using a Mechanical pencil or an exact-o knife fixing CPU pins....

socket7/370 a mechanical pencil is the best method. although an exacto or razer is easiest to get them into rows.

Contest to give away the damaged board for someone to try in desperation to get the pins straightened?
I would be game....
 

Pusher of Buttons

[H]ard|Gawd
Joined
Dec 6, 2016
Messages
1,924
How many hours have I spent using a Mechanical pencil or an exact-o knife fixing CPU pins....

socket7/370 a mechanical pencil is the best method. although an exacto or razer is easiest to get them into rows.


I would be game....

I've never wrecked pins I couldn't repair on a PGA but LGA I've scrapped a few
 

toast0

2[H]4U
Joined
Jan 26, 2010
Messages
2,162
I understand why they did it that way, but it underscores just how badly we need a new mounting method in the industry. Optical interconnects or something

So many of those pins are for ram. Switching ram from parallel to serial would reduce pin count, but increase latency; regardless of optical or not. Unless you're going to somehow multiplex everything into a small number of optical interconnects, it's going to be extra fiddly too. An alternative, probably terrible way to avoid all these pins, would be a pluggable 'cpu module' with cpu soldered on to a board with the ram slots, and then PCI-E and power to the motherboard.
 

Gasaraki_

Gawd
Joined
Oct 27, 2016
Messages
617
I understand why they did it that way, but it underscores just how badly we need a new mounting method in the industry. Optical interconnects or something. I know you messed up, you admit you did and the motherboard maker has nothing to do with it. It's AMD's standard. But that really is kind of lunacy on pin count and square inches.

We've had good luck since the industry started to put the pins on the motherboards. None of us knew how that would work out in the beginning. It stopped the rampant pin damage on expensive CPUs. And mostly in enthusiast situations the motherboard is a less expensive item to destroy than the CPU.

But that socket is so large only people who trust themselves, have steady hands and know PC builds inside out should be screwing with that socket at all. (You certainly do, and an accident still happened) Even small debris contaminates in the pin area are an issue with that thing. You really wouldn't want to leave that pad exposed for very long. It's like a Murphy's Law Magnet. There's no way he can resist causing everything to go wrong. :p

Intel LGA "pins" you can push down all you want. They seem pretty good better than pins on a cpu.
 

tangoseal

[H]F Junkie
Joined
Dec 18, 2010
Messages
9,426
I have to admit that Kyle is an friggin hero for making these mistakes. Think of the boards saved when we get ours because he was the great sacrifice and let us know just how much caution we need. I plan to install my 1950x once and only once and if I sell it that will be the only time I remove it.
 

steakman1971

2[H]4U
Joined
Nov 22, 2005
Messages
2,433
I installed an AMD CPU a few months ago and somehow bent a pin. I got out my Exacto knife and mechanical pencil to straighten the pins. Couldn't believe I did it - I thought I was being careful. The CPU powered up and has been running solid since my repair work. (I also had to get my reading glasses out to see the damn pins - I wouldn't want to try to do this with smaller pins. My eyes aren't that good (and shakey hands don't help)
 

JimmiG

2[H]4U
Joined
Apr 3, 2008
Messages
2,429
So many of those pins are for ram. Switching ram from parallel to serial would reduce pin count, but increase latency; regardless of optical or not. Unless you're going to somehow multiplex everything into a small number of optical interconnects, it's going to be extra fiddly too. An alternative, probably terrible way to avoid all these pins, would be a pluggable 'cpu module' with cpu soldered on to a board with the ram slots, and then PCI-E and power to the motherboard.

How come Skylake X only has 2066 pins? It also has quad-channel memory support and up to 18 cores. It has slightly fewer PCI-E lanes but could that account for a difference of nearly 2000 pins?
 

RogueTadhg

[H]ard|Gawd
Joined
Dec 14, 2011
Messages
1,527
[H]ardocp could make some good posters with hires pictures like this.
 
Last edited:

Silentbob343

[H]ard|Gawd
Joined
Aug 2, 2004
Messages
1,943
Whatever you do keep q-tips away from sockets, they will get snagged and bend pins like a motherfucker...or so I heard from a friend.
 

tangoseal

[H]F Junkie
Joined
Dec 18, 2010
Messages
9,426
How come Skylake X only has 2066 pins? It also has quad-channel memory support and up to 18 cores. It has slightly fewer PCI-E lanes but could that account for a difference of nearly 2000 pins?
64 pcie lanes is a shit ton and there is alot of electrical connectivity to make them interface the CPU.

Also consider that the sp3 socket is a server socket where there are two more functional dies that give the Epyc 32 cores. There is many many pins that support the two other dies. Amd didnt want to make another socket for these as Threadripper is essentially either a failed Epyc or a Twin Ryzen in an Epyc frame. Im sure well know for sure in future.

Im.seriously hoping they are failed epycs that we can unlock more cores on. Id love a 20 or 24 core CPU hah
 
Last edited:

MV75

[H]ard|Gawd
Joined
Nov 13, 2007
Messages
1,025
Those cpu's are frigging huge. Did they just find a new purpose for some old Pentium pros? :D

I see the return of the secc a thing.
 

erexx

Limp Gawd
Joined
Jul 14, 2004
Messages
472
Can we please get a pic of the damaged part?
Would be good to see exactly how the pins are bent because it hard to tell exactly what the geography of the pins are from the pic of the good socket.
 

JustReason

razor1 is my Lover
Joined
Oct 31, 2015
Messages
2,483
64 pcie lanes is a shit ton and there is alot of electrical connectivity to make them interface the CPU.

Also consider that the sp3 socket is a server socket where there are two more functional dies that give the Epyc 32 cores. There is many many pins that support the two other dies. Amd didnt want to make another socket for these as Threadripper is essentially either a failed Epyc or a Twin Ryzen in an Epyc frame. Im sure well know for sure in future.

Im.seriously hoping they are failed epycs that we can unlock more cores on. Id love a 20 or 24 core CPU hah
Would there be FAILED Epyc chips though? Wouldn't they already know if a die (8core) is fail before putting on the interposer?
 

tangoseal

[H]F Junkie
Joined
Dec 18, 2010
Messages
9,426
Would there be FAILED Epyc chips though? Wouldn't they already know if a die (8core) is fail before putting on the interposer?
Besides what Dr8Baeur unconvered I am not sure we know much at all about these chips (yet).

Yes they should and would. To my understanding there are 4 dies under there. Now AMD is "claiming" they are none but spacers under there but I smell something else. I smell two failed or one failed die of the 4 total. They didn't make the cut as 32 core Epycs. I have no idea though and I do not know a large amount about processor fabrication.

There is something about them that AMD wanted hidden as they requested Dr8, or however you spell his gaming name, take them down immediately off his site.

As far as the interposer, well that depends if they ran all the "wires" so to say. If they didnt run wires to the other two dies then there is no physical way to enable cores like past chips. But if wires are run then that is where my knowledge doesn't continue. If these are not failed epycs then why use the same exact socket etc.... for these Threadripper chips? Seems to me AMD had quite a few that didn't make the cut so they are like hrmmmm Threadripper. Just a hunch, again, probably wrong as I am with everything in life.
 
Last edited:

Gman1979

Gawd
Joined
Sep 21, 2011
Messages
640
Meh.10 MG of Diazepam, very good lighting, and quality magnification can have her back in fighting shape in a couple hours. I just worked this trick on a 1150 board I wrecked some pins on when I first built my signature rig. Diazepam reduces the natural twitching and shaking you get doing things requiring very fine motor control such as messing with tiny ass #$@^&*% pins. My father mentioned using it for the very same purposes when he was a sniper, though only for range use or competitions. The tiniest movement when firing a rifle can turn into a much larger deviation at 1,000+ yards :D , much like a tiny twitch can send your tool of choice wrecking into pins you never meant to touch as well.
 

Drewis

Limp Gawd
Joined
Jul 25, 2006
Messages
453
Dont feel bad Kyle. Been doing this awhile too and after 20 years dropped a x5680 directly, corner down, on the 1366 socket. Mashed 6 pins to shit. Even though I was drinking at the time I still didnt have steady enough hands and fat fingered it as well.

Took 1 hour and she booted up though, unlike this. That socket seems insane.. Board looks amazing though!

Looking forward to the reviews and thanks for the PSA.
 

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
54,075
Can we please get a pic of the damaged part?
Would be good to see exactly how the pins are bent because it hard to tell exactly what the geography of the pins are from the pic of the good socket.
Sorry, board is gone, back to GBT.
 

mnewxcv

[H]F Junkie
Joined
Mar 4, 2007
Messages
8,868
I don't think I'm the only one that has ripped a ryzen cpu out of the socket when removing the hsf. I really like the idea of a more secure mechanism to hold down the cpu. Fact is, Intel has done this for generations. Glad amd is following, but hope it makes it down to the desktop level not just hedt.
 

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
54,075
I don't think I'm the only one that has ripped a ryzen cpu out of the socket when removing the hsf. I really like the idea of a more secure mechanism to hold down the cpu. Fact is, Intel has done this for generations. Glad amd is following, but hope it makes it down to the desktop level not just hedt.
You would be correct sir.
 
Top